Mini Post #1: The Looming Gay Housing Crash
Mini Post #1: The Looming Gay Housing Crash
Do gay couples face discrimination in the housing market?
I’ve decided to try something new: once a week, to write a shorter, narrower post focusing on one specific paper. I’ll find a tab to put them all in and link it here.
Onto the actual post: do gay couples face discrimination in the housing market?
Discrimination against gay people, aka homophobia, is real (citation needed). Most studies of anti-LGBT discrimination focus on the labor market, but at least one paper focuses on the housing market: Nir Eilam and Yeonjoon Lee’s “New Results on the Disparities between Same-Sex and Different-Sex Couples in the Home Mortgage Market”, from this year (2024).
The question is interesting because, as many people might imagine, owning a home is important to people. Firstly, because it is the most valuable asset most households own, meaning that differentiated access would result in radically different wealth trajectories. Secondly, there are cultural factors (in the US) that show homeownership as a sign of personal and financial success. So, whether this “American Dream” of a white picket fence is accessible to gay people as well, and what the differences are in case there are any, is an empirically interesting question.
Well, there’s new results: what about the old ones? Previous papers (for example, this one) found that there are significant discrepancies between the rate at which gay couples and straight couples are accepted, or rejected, for mortgages. In another study, using a randomized survey from the Boston Fed, it was found that gay people who applied for mortgages in 1990 wre 73% more likely to be denied the application than similar straight people - compared to a difference of 40% for racial minorities vs white applicants with similar characteristics. Obviously, this is a very small sample from one city from 34 years ago, but it can answer the question in the affirmative.
What do we mean when we say “similar straight people”? It means that those factors are controlled for. For example, imagine if you wanted to test whether players from Bolivia are more skilled at basketball or have better training than players from Denmark, by tallying up how many points (the basketball equivalent of goals) each country scored in international competitions. One first possibility is that, unrelated to each country’s skill level, Denmark is much richer, meaning that the government might finance more participation in tournaments - boosting their stats, which would have to be adjusted for. Additionally, Denmark is one of the tallest nations on Earth, while Bolivia is one of the shortest - so if you think it’s interesting to know the role of “training” or “skill” instead of pure height, you could control for height.
The most important part is that what you’re trying to estimate is the impact of an unobservable factor, not an observable one. If you wanted to measure the impact of height, you would simply run a regression and statistical tests on height directly. But to verify if an unobservable is present, you have to isolate it from observables - which does not preclude them from being important. In the previous example, height is obviously an important factor for basketball, but that’s not the question we were asking. Similarly, claims that non-white or female pilots perform as well as male, white pilots controlling for experience are valid as long as you’re looking for explanations that focus gender and race, not experience, as the source of their performance.
Anyways, back to gay people’s mortgages. Comparing apples to apples is important because, in the mortgage market, many qualities are very important: income, size of down payment, education, kind of job performed, what kind of housing is purchased, neighbourhood, etc. A lot of these qualities are linked to labor market outcomes, which are also linked to gayness, so it would be comparing the average Bolivian to the average Dane on a hoop competition - not great. Controls can be useful, thus, if minding that each specific one can fail, both by improper measurement, or by correlation to the variable in question.1
A second, more improtant issue is that not all LGBT people are open while identifying as such: many don’t, because of personal reasons or discrimination, while others do. If the ones who are open about being gay are systematically different across variables of interest (income, education, partnerships, etc.) than the ones who aren’t, then comparing the openly gay to the straights will result in incredibly biased comparisons. The way to solve this is twofold: first, you could use some direct indicator, such as declared partners, from the disclosures people make when applying for a mortgage. Since that data is very private, it’s not always available - though it was for the paper in question. A second way is to use data that is actually available, such as anonymized indicators of the person co-signing or guaranteeing the loan, and making some assumptions: if one of the two co-applicants is significantly older than the other one they’re probably a parent and chiled, while unrelated people who cosign a mortgage are usually a couple.
The fact that this paper has access to confidential, anonymized data is important because it means that the research can also control for other variables of interest, such as credit scores or financing conditions, that might affect the outcomes of the loan application process - maybe only gay people with substantially higher credit scores, controlling for income, apply for mortgages in the first place. Other relevant factors could be age, since gay people may simply be younger than straight people, or gay couples may be whiter than straight couples (at least, gays applying for mortgages), or other factors that are harder to disentangle.
One last consideration is that the market for mortgages has a lot of information that is available to the applicant but not to the bank, and which the borrower has no reason to disclose - for example, about their broader financial prospects, or personal financial habits. Additionally, since not everyone who applies for a loan can actually pay it off, banks need careful vetting of applicants, particularly since carelessness could attrack worse, on average, applicants than carefulness - “the bank with no standards” would only get borrowers who can’t get a loan anywhere else. To make up for this lack of data, banks tend to make inferences about consumers based on the data they do have, such as the neighborhood the house is located. This is important because it is possible that gay couples aim to purchase homes in neighborhoods that banks do not feel comfortable issuing mortgages for, or housing with amenities they do not think wise to finance. A further reason could be that gay people tend to live in larger cities, while straight people are spread out all over the place, which would mean that “big city” specific effects play a role in potentially higher denial of mortgages. So specific controls for the effect of neighborhood, region, and city (as well as year - who’s getting approved for a mortgage in 2009. straigth or gay?) are in order.
With all that said, is there evidence of discrimination? There isn’t really any for applications where the main applicant is a woman and the co-applicant is her husband, for instance. But for gay couples? Yes. Compared to a straight person of similar economic characteristics, a gay male couple is 8.8% more likely to be denied a mortgage than a straight couple, and are on average charged 0.035% more for their mortgages, i.e. an increase of 0.8%. For lesbian couples, compared to applications where the main applicant is a woman, the interest rate effect is a fifth smaller than the gay male effect (0.028%, or 0.64%), and women who apply for mortages also face a small interest rate penalty (around a third of the penalty for gay men) vs men who apply, but there is no significant difference in mortgage rejection rates - with either (straight) women or straight men. There are no disparities whatsoever in the mortgage refinancing market. These disparities worsened during the COVID and post-COVID period, since the sample ranges from 2018 to to 2021.
There are also heterogeneities by region: while being gay in the Northeast and the Western states, which are generally the most progressive, has no particular impact on mortgage application rates or interest rates (i.e., the results are in line with the national results outlined above), applying for a mortgage in the Midwest as a gay man has a 22.7% higher likelihood of rejection than for a similar straight man in the same region, while the South (the most conservative region) has a 10.3% increase in likelihood of mortgage rejection. By the level of acceptance towards gay marriage in each tertile of states (i.e. third, or 16.666 out of 50 states), gay men in the 17 most conservative US states don’t have much of a difference with the rest of the country, and neither do gay men in the most progressive third of states; the difference manifests precisely in the middle third of the state LGBT acceptance distribution. These results do not extend to lesbian couples, whose outcomes are broadly in line with other female-led applications (who are, in turn, in line with male applications).
What could this mean? Well, let’s start with the gay male result: there is clear evidence of an unexplained disparity in outcomes, which might be called discrimination. The regional discrepancies, meanwhile, make a perverse kind of sense: since the most conservative states have the fewest gay people wanting to permanently settle in them, they might be markedly different from other gays in some observable or unobservable manner, resulting in applications so different that it offsets the effect of homophobia. In contrast, the most progressive states simply are not homophobic enough for divergent application outcomes. In the middle of the road (both in terms of acceptance, but also politically - the Midwest is very competitive), there are enough “normal” gays for the existing, “normal” homophobia to actually affect them.
For the lesbian result, the lack of discrimination vis a vis straight women (this comparison is made in case there is specific gender-based discrimination, or some other gendered effect) is weirder. It could be due to the fact that lesbians have, on average, higher incomes than similar straight women, but lower incomes than straight men - meaning that, if there is a relative “weakness” from being a lesbian, it could be counteracted by this “strength”. The interest rate penalty, however, is hard to square with this fact, so it’s possible there is some weaker and more heterogenous penalty.
But the most surprising result, by far, is the disparity in mortgage delinquency rates: gay male couples have a 53.9% higher chance of defaulting on their mortgages than straight couples - 3.81% for straights, versus 5.86% for gays. There is a similar effect for lesbians, which is significant but a third of the size (4.59%, a rise of 20.5%). It could point towards an unobserved instability in gay relationships, that make loaning to them riskier - for instance, higher separation rates, higher exposure to employment shocks (i.e. getting fired), or lower parental support. Exactly why this kind of lurking instability exists is kind of puzzling, considering that, for instance, gay men have significantly lower divorce rates than lesbians, whose rate of separation is sky-high2.
To finish up, some links
The original paper, by Eilam and Lee
A related paper by Sun and Gao, with similar results and some key differences.
A previous blog post on the economics of the LGBT community
A previous blog post on the limits and dangers of controlling for observables
To elaborate a bit before stupid points vis a vis the wage gap get made: controlling for observables helps establish their own causal impact, as well as clarify the role that unobservables play - for example, if the unobserved component of the wage gap was large, it would point to explicit discrimination, or a large difference in preferences. This doesn’t mean that discrimination, or the expectation thereof, cannot affect the specific observables (career choice, work experience, hours worked).
To use a simpler example: this paper finds that women and men in economics have equal rates of publishing papers, controlling for them receiving tenure. Far from being a sign of gender equality, it is an indictment: women, especially mothers, were far less likely to receive tenure in the first place, selecting the “best” ones for the comparison.
The main reason, per most studies, appears to be that a lot of lesbians got married really soon, especially in the mid-to-late 2010s when it was freshly legalizd, and like most quick marriages, they ended in divorce. “My wife left me” is a cross-sexuality sentiment, it seems.
One thing that affects bank mortgage availability is the size of the property- not just its location or high end appeal. Single bedroom places, the old bungalow in Hollywood or conversely the 1bedroom flat in Wilmington (CA) both have to be financed with cash. No bank loans or bank money available (except your credit card).